Edition: February/April 2018


Overhyped or a revolution in asset management?

Hywel George, director of investments at Old Mutual Investment Group, offers an answer.

2017 saw the advent of the first fully AI-powered, daily traded exchange-traded funds. Some people view this as heralding a shift into a new investment paradigm, Autonomous Learning Investment Strategies (ALIS).

George . . . human combination

What’s new about these investment processes is that the technology learns and adapts as it goes along, based on the information and enormous data sets to which the algorithms have access. It’s on these that they base their investment decisions and solve problems, all without human input.

As in other AI fields, it has raised the spectre of ’singularity’ – a much-vaunted future state when computers could potentially have superintelligence that surpasses our own and which could, it is feared, ultimately put humans out of business. But have we truly crossed the AI Rubicon or is this merely hype?

For many, the AI milestones achieved over the past five years have set us up for the greatest technological revolution in history over the next decade. The investment industry will undoubtedly be at the centre of it. (See table).

But artificial intelligence and talk of technological revolution have been around for a while. For instance, in 1965 when a British mathematician and cryptologist brought up the concept of an intelligence explosion. Then, in 1993, a sci-fi writer and computer scientist predicted that within 30 years we would have the means to create superhuman intelligence.

There are many definitions of AI but Forbes magazine contributor David Thomas put it succinctly: Artificial intelligence is a branch of computer science that aims to create intelligent machines which teach themselves.

There are different levels of AI. Each level becomes more sophisticated and autonomous in the tasks computers can do without human intervention. There is machine learning (or structured learning) which is the ability of computers to learn and improve at tasks with experience. Then there is deep (or unstructured) learning, when a computer uses algorithms that adapt to new data and thus trains itself to perform tasks. The best-known examples of deep learning are IBM Watson and driverless cars. (See graphic).

A deeper understanding of AI

Timeline of AI milestones

The first Dartmouth College summer AI conference is organized by John McCarthy, Marvin Minsky, Nathan Rochester of IBM and Claude Shannon.
Joseph Weizenbaum (MIT) builds ELIZA, an interactive program that carries on a dialogue in
English language on any topic.
Herbert A Simon wins the Nobel Prize in Economics for his theory of bounded rationality, a cornerstone of AI known as “satisficing”.
Vernor Vinge publishes “The Coming Technological Singularity,” predicting that, within the next 30 years, humankind would have the ability to create superhuman intelligence.
The Deep Blue chess machine (IBM) defeats the (then) world chess champion, Garry Kasparov.
Vernor Vinge publishes “The Coming Technological Singularity,” predicting that, within the next 30 years, humankind would have the ability to create superhuman intelligence.
The Deep Blue chess machine (IBM) defeats the (then) world chess champion, Garry Kasparov.
Google builds self-driving car.
IBM’s Watson computer defeated television game show Jeopardy! champions Rutter and Jennings.
Google DeepMind’s AlphaGo defeats 3x European Go champion Fan Hui by 5 games to 0.
Google’s AlphaGo Zero - an improved version of AlphaGo - learns by playing only against itself and
beat its predecessor 89:11 after only 40 days Source: Wikipedia, Old Mutual Investment Group.
Source: Wikipedia, Old Mutual Investment Group.

Inevitably, the advances in AI have spurred robust debate about what impact AI will have on the investment world. To get a balanced perspective, it’s worth considering why AI is developing so rapidly.

AI advances have primarily been made possible by the sharp decline in the price of graphics processing units (GPUs) in recent years. Driven by gaming, it has enabled AI to access immense amounts of data of all types (numerical, image, voice) being made available from companies such as Google, Facebook and Microsoft.

Cloud-based hosting has also provided access to extremely strong AI platforms. For instance, you can use IBM’s or Google’s AI platforms to take advantage of work that they have already done and build on top of this.

Why is this important?

  • It allows for quick-to-market implementation when you have sufficient data to teach your algorithm;
  • With so much data being made available, you don’t even need to come up with an hypothesis to code in. You can throw mountains of data at the AI and, through deep learning, it will figure out the pattern;
  • Platforms are cheap or free, so the barriers to entry are low. The main barrier is access to sufficient rich data.

Notwithstanding the increasingly fast-paced innovation we’ve seen, and the growing excitement about the potential of AI, it is not likely to be an investment panacea. It’s premature to think that fundamental qualitative investment professionals will no longer have jobs as a result of AI.

Instead, some of the things the investment industry needs to be thinking about are:

  • If you pick the incorrect data, you will get the incorrect result (which will come in as the correct result, but is based on the wrong information);
  • An algorithm learns as time goes by, but it cannot determine an upcoming black swan event unless it has a previous black swan event from which to have learnt;
  • AI is good at doing one thing well, but not at integrating many things into a ‘super-solution’. For instance, you can use AI to determine what the market may do using machine readable news as a factor in an investment portfolio. But you are not able simply to ‘ask AI to come up with a portfolio’ and let it figure out how to do so.

More important for the investment industry is to consider how can we use AI to improve portfolios and remove from our jobs the repetitive grudge aspects so that we can concentrate more time on the hard-thinking work; in other words, how we use AI to augment what we do as opposed to worrying about it replacing what we do.

This is not a matter of human versus machine but of human and machine being better than human alone.